Protecting Language Generation Models via Invisible Watermarking
This addresses the problem of intellectual property protection for model providers against distillation-based theft, representing an incremental advance over previous watermarking techniques.
The paper tackles the problem of protecting language generation models from model extraction attacks via distillation by proposing GINSEW, a method that injects invisible watermarks into probability vectors during decoding, resulting in an absolute improvement of 19 to 29 points in mean average precision for detecting infringement with minimal impact on generation quality.
Language generation models have been an increasingly powerful enabler for many applications. Many such models offer free or affordable API access, which makes them potentially vulnerable to model extraction attacks through distillation. To protect intellectual property (IP) and ensure fair use of these models, various techniques such as lexical watermarking and synonym replacement have been proposed. However, these methods can be nullified by obvious countermeasures such as "synonym randomization". To address this issue, we propose GINSEW, a novel method to protect text generation models from being stolen through distillation. The key idea of our method is to inject secret signals into the probability vector of the decoding steps for each target token. We can then detect the secret message by probing a suspect model to tell if it is distilled from the protected one. Experimental results show that GINSEW can effectively identify instances of IP infringement with minimal impact on the generation quality of protected APIs. Our method demonstrates an absolute improvement of 19 to 29 points on mean average precision (mAP) in detecting suspects compared to previous methods against watermark removal attacks.